Scholarlink

<p>Atmospheric particles significantly degrade image quality, posing challenges for computer vision applications requiring high-contrast and clear images. To address this issue, a novel parallel architecture for real-time image dehazing is proposed, incorporating an optimized atmospheric light estimation module and an enhanced guided filter for transmission coefficient estimation, designed for hardware-efficient implementation. Extensive qualitative and quantitative comparisons were conducted across Raspberry Pi Model 4B, PYNQ Z2, and NVIDIA Jetson Nano, through which the proposed parallel architecture was shown to outperform existing techniques in both processing speed and dehazing quality. The PYNQ Z2 platform was found to deliver superior perceptual and structural performance across all evaluation metrics, including PSNR, SSIM, MS-SSIM, CORR, MSE, and LPIPS(d). The effectiveness of hardware-level, parallel dehazing for enhancing image visibility and system reliability under adverse conditions is thus demonstrated. The source code and datasets used in this study have been made publicly available to support reproducibility and further research.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Real-Time Image Dehazing via a Novel Parallel Architecture: An Empirical Comparison of Hardware Platforms

  • 期刊论文

  • 发表日期：

    2025年5月28日

查看全文
  • Avra Ghosh,
  • Sumanta Bhowmic,
  • Sheli Sinha Chaudhuri

摘要

Atmospheric particles significantly degrade image quality, posing challenges for computer vision applications requiring high-contrast and clear images. To address this issue, a novel parallel architecture for real-time image dehazing is proposed, incorporating an optimized atmospheric light estimation module and an enhanced guided filter for transmission coefficient estimation, designed for hardware-efficient implementation. Extensive qualitative and quantitative comparisons were conducted across Raspberry Pi Model 4B, PYNQ Z2, and NVIDIA Jetson Nano, through which the proposed parallel architecture was shown to outperform existing techniques in both processing speed and dehazing quality. The PYNQ Z2 platform was found to deliver superior perceptual and structural performance across all evaluation metrics, including PSNR, SSIM, MS-SSIM, CORR, MSE, and LPIPS(d). The effectiveness of hardware-level, parallel dehazing for enhancing image visibility and system reliability under adverse conditions is thus demonstrated. The source code and datasets used in this study have been made publicly available to support reproducibility and further research.